On the Security Vulnerabilities of Text-to-SQL Models